From safe harbours to AI harbours: Reimagining DMCA immunity for the generative AI era

Abstract

Generative artificial intelligence (AI) overturns the passive-intermediary assumptions that underlie the US Digital Millennium Copyright Act (DMCA) safe harbour. Modern systems ingest vast, often unlicensed datasets and emit on-the-fly outputs through a supply chain that spans data suppliers, model developers and deployers - raising parallel concerns in the EU, UK, Hong Kong and other jurisdictions. Building on DMCA section 512, this article sketches an 'AI harbour' that ties immunity to role-specific duties: provenance disclosure and transparency for data suppliers; dataset curation, memorization-mitigation and watermarking for developers and dynamic filtering, complaint handling and repeat-infringer policies for deployers. A new statutory section - administered by an 'AI Division' within the Copyright Office - would certify actors, audit compliance and endorse technical standards developed through industry co-regulation. The proposal preserves the DMCA's cooperative bargain while supplying clear, technologically realistic compliance pathways. Because its tiered obligations, administrative oversight and adaptive self-regulation can be grafted onto existing regimes, the model travels well: the EU could integrate comparable safeguards alongside the Digital Services Act and AI Act; the UK's post-Brexit reforms and Hong Kong's technology-neutral Copyright Ordinance could embed similar structures. In this way, the AI harbour could offer a scalable blueprint for protecting creators without chilling innovation in the generative era.